#27 · ML / AI methods

Agents chain target discovery, pocket design, and PK simulation for therapeutic design

The dominant thread is the shift from single-model property prediction to agentic LLM systems that chain biological tools toward target and therapeutic decisions. OrchestRA couples knowledge-driven agents with human-in-the-loop control across target discovery, pocket-based design, and PBPK pharmacokinetic simulation in an iterative loop. Alvessa applies an evidence-grounded agentic assistant to functional genomics and drug target assessment, verifying its statements against retrieved records. What is new is the emphasis on verification and auditability alongside generation, rather than prediction accuracy alone. SpatialBench contributes a complementary perspective, testing whether agents recover established results from real spatial transcriptomics workflows across 146 verifiable problems. It is a benchmark rather than a discovery system, so it evaluates the same agentic paradigm rather than drug-target prediction directly.